GIS Partial Discharge Development Process Research And Its Severity Evaluation Based On The Multiple Parameters

Partial discharge (PD) is the external manifestation of the internal insulation degradation in gas insulated switchgear(GIS).It has an important significance for GIS safe operation to realize its severity evaluation.Based on this, this article is in full consideration of the common insulation defects in GIS and its actual operating conditions, and then the typical partial discharge development process and its severity was studied. The main achievements are as follows:Firstly,the 252kV GIS experiment platform was set up and four typical partial discharge models were established combined with the common insulation defects in GIS. The development process of partial discharge caused by these typical insulation defects has been tested for a long time by using ultra high frequency(UHF) method,ultrasonic method and ultraviolet imaging method. And got the GIS internal partial discharge characteristic information,then got a variety of statistical spectrograms of different discharge stages,and take these statistical spectrograms as the basis of GIS partial discharge severity classification.The dimension of feature space got from experimental feature spectrograms is high,principal component analysis is a kind of classic dimensionality reduction method,it can realize feature space dimension reduction,but after the dimension reduction,it can not explain the relations between principal components and characteristic parameters.Based on this,the article introduced non-negative sparse principal component method,results showed that this method could achieve low-dimensional features extraction,and the clustering result of the components had been enhanced.At present,mainly take the characteristics comes from a single detection method to realize partial discharge pattern recognition,in order to effectively use the characteristics from different detection methods,the article adopts Multi-Kernel Learning Relevance Vector Machine(MKL-RVM)algorithm that can blend characteristic parameters from different detection methods,and then optimized kernel function parameter based on K-fold cross-validation and Particle Swarm Optimization(PSO) methods,which improves the performance of MKL-RVM,finally blended the UHF and ultrasonic characteristic parameters,and proved the effectiveness of this method by an example.Finally,according to the different experimental phenomenon and the corresponding statistical characteristic of the four kinds of typical partial discharge,the partial discharge development process was divided into slight discharge,moderate discharge and serious discharge stage,using UHF discharge amplitude,discharge frequency and ultrasonic signals equivalent root-mean-square bandwidth as discharge severity evaluation parameters,using the analytic hierarchy process(AHP) to obtain the weight of parameters,and finally established the membership functions of the four kinds of discharge parameters,and fuzzy comprehensive evaluation was adopted to realize the GIS partial discharge severity evaluation,and it showed that the method is scientific through the experiment.